Research On Detection Of Lung Nodules And Classification Of Benign And Malignant Lungs Based On CNN

According to the "Global Cancer Statistics 2018" released by the World Health Organization,the prevalence and mortality rate of lung cancer ranks first among all types of cancer,which seriously affects human health.Clinical studies have shown that if lung cancer is treated early,the five-year survival rate of lung cancer patients will reach more than 80%.Therefore,early screening of lung cancer is of great significance for improving the survival rate of patients.In clinical diagnosis,doctors need to accurately diagnose the location and benign and malignant lung nodules.Existing medical image-assisted diagnosis methods include chest X-ray film,computed tomography imaging,positron emission computed tomography,and magnetic resonance imaging.Due to the advantages of high resolution and low price of CT technology,doctors usually use CT imaging technology to diagnose patients.However,with the rapid increase in lung cancer patients in recent years,it has increased the workload of imaging physicians.With the development of deep learning and computer technology,it is possible to use artificial intelligence to assist doctors in diagnosis.This paper analyzes the shortcomings of existing lung nodule detection and benign and malignant classification,and proposes multi-scale fusion convolutional neural network-based lung nodule detection and lung conjunctival benign and malignant classification.The main research contents are as follows:(1)This paper presents a lung nodule detection algorithm based on a multiscale fusion convolutional neural network.Because the small lung nodules and blood vessels in CT images are very similar in shape,size and gray scale,it is easy to cause misjudgment and missed judgment in clinical diagnosis.To solve this problem,this paper first constructs a multi-scale convolutional neural network to obtain the shape and texture features of lung nodules.On this basis,in order to improve the detection of small nodules,shape features and texture features are fused to obtain a multi-scale fused feature map,and the fused feature map is used to detect small nodules.In addition,multi-scale feature maps are used to detect lung nodules of other sizes.Compared with the existing lung nodule detection algorithms,the recall rate of the proposed method is 96.8%,which is 4.94% higher than the existing methods.(2)A benign and malignant classification algorithm for lung nodules based on a full convolutional neural network is proposed.Traditional convolutional neural networks consist of convolutional layers,activation functions,pooling layers,and fully connected layers.When benign and malignant classification of lung nodules,the pooling layer will cause a lot of information loss,and the fully connected layer will make the neural network parameters redundant.These two problems of traditional neural networks affect the classification performance of benign and malignant pulmonary nodules.Therefore,this paper addresses the problem of the pooling layer by using a convolution layer with a step size of 2 instead of the pooling layer.In order to reduce the parameter redundancy of the fully connected layer,a global mean pooling layer is constructed instead of the fully connected layer.Through experiments,the accuracy,recall,and AUC values of the algorithm reached 93%,92.7%,and 0.92,respectively.